Methodology
How SaferClaw Works
SaferClaw is a capability-based risk index for skills and tools in the OpenClaw ecosystem. It is designed to quickly answer what a tool can do before you enable it.
What the index measures
Every tool is scored on three dimensions using deterministic evidence.
Locality
Where execution happens.
- Local: local CLI only.
- Hybrid: local plus remote API.
- Cloud: hosted service.
Data access
What kind of data can be read.
- Public: docs and public pages.
- Personal: calendars or profile data.
- Sensitive: secrets and tokens.
Actions
What the tool can change.
- Read: query-only operations.
- Write: editing files/settings.
- Execute: shell commands/deployments.
How rating is computed
- Max-risk rule: if any detected capability is higher, that dimension is bumped to the higher level.
- Hybrid locality: the tool mixes local execution and cloud services (local+cloud).
- Each detail page includes evidence links and bullet explanations for why each level was assigned.
How to use this index
- Use it to triage tools before install or enablement.
- Use it to compare alternatives with similar features but lower capability risk.
- Open each tool detail page and follow the recommended Best practices before enabling it.
- Use it to set approval thresholds for personal devices, team workspaces, and production environments.
Limitations and disclaimer
Methodology and disclaimer
- This is not a full security audit and not legal advice.
- Some source data can be incomplete, stale, or missing context.
- Skills and tools are sourced from ClawHub through automated fetches.
- Always validate permissions and behavior before enabling tools in production or high-stakes workflows.
- Important: this system is built for fast coverage, not exhaustive analysis. We do not run full security scans on every skill.
- Ratings are provided in good faith based on available signals, but they should always be independently verified. No guarantees are made.
Sources and update cadence
- Ratings are generated from skill metadata, capability signals, and published evidence links.
- Index data refreshes when the ingest and rating generation pipeline runs.
- Per-tool pages expose source links used for the current snapshot.
Contribute or report issues
Found a wrong rating or stale evidence? Send a report with the tool slug and what looks incorrect. Seriously, just message me or reply to me on X.